Feature Point Matching Algorithm Based On Fish-eye Image

Fish-eye lens has the characteristics of super wide Angle,which can absorb more information than ordinary lens.Therefore,the fish-eye lens can cover the whole 3D space with a small amount of images,which is widely used in the fields of driverless cars,robot navigation and positioning,geographic information system and so on.Among them,image feature point matching technology is one of the core technologies in the above applied fields.The nonlinear distortion is also introduced into the fish-eye image to obtain large wide-angle information,which seriously affects the matching quantity and accuracy.In order to solve this problem,this paper proposes the feature point matching algorithm based on curve descriptor and local(Descriptor-Nets,D-Nets),and the feature point matching algorithm based on triangular mesh and(Affine-SIFT,ASIFT).The main research work is as follows:First of all,starting from the feature point matching of ordinary perspective images,this paper analyzes the differences between ordinary perspective lenses and fish-eye lenses in the imaging model,introduces the basic principle of feature point matching,analyzes the performance of existing feature point matching algorithms,and puts forward the reasons why these algorithms are not applicable in fish-eye images.Secondly,aiming at the influence of nonlinear distortion of fish-eye image on feature matching descriptors,a matching algorithm based on curve descriptors and local D-Nets is proposed.The D-Nets algorithm is a robust affine image matching algorithm.Since the line descriptor of the D-Nets algorithm is not applicable to fish-eye image,a curve descriptor based on the hemispherical model of fish-eye image is proposed.At the same time,aiming at the problem of high computational complexity and slow matching speed of D-Nets algorithm,this paper improves the traditional global D-Nets matching algorithm to local D-Nets matching algorithm.Experimental results show that the proposed method can obtain more accurate matching points than the global D-Nets algorithm.Then,an improved affine image matching algorithm,ASIFT,is proposed based on triangular mesh and ASIFT.This algorithm used the local affine model to transformfish-eye image matching into multi-region affine image matching.The fish-eye image is divided into several image regions by constructing a triangular mesh,and local affine matching is adopted for each region.Then,based on the idea of motion smoothing statistics,the operation of removing mismatches is carried out on the basis of the triangular mesh.This algorithm can obtain a large number of matching points in fish-eye image matching,and can quickly distinguish the correct match from the wrong match,effectively improving the stability and accuracy of fish-eye image matching.Finally,we carry out feature point matching experiments on the images of zoom,translation and affine transformation in different scenes.The experimental results show that the matching algorithm based on curve descriptor and local D-Nets achieves the lowest error matching rate,and the matching points are evenly distributed.The matching algorithm based on triangulated mesh and ASIFT obtains the most matching points in all algorithms.